Anthropic’s Mythos 5 Wins Partial US Approval, Fable 5 Stays Locked

The US Commerce Department has cleared Anthropic to redeploy its Mythos 5 cybersecurity model to a narrow list of about 100 federal agencies and private companies, ending a two-week standoff that began when the government pulled both of the company’s newest AI models offline over national security concerns. Commerce Secretary Howard Lutnick’s Friday letter to Anthropic’s chief compute officer Tom Brown said safeguards are now adequate for a curated set of trusted partners, and that Anthropic can restart access for that group.

The same letter leaves Fable 5, the public version of the same underlying model, still blocked. Hours before Lutnick signed, OpenAI launched its GPT-5.6 family of three models under the same ad-hoc government vetting process Anthropic just navigated, a sign that the Friday decision is the start of a wider precedent for frontier rollouts in Washington.

What the Letter Says, and What It Doesn’t

Lutnick’s determination, in the letter viewed by CNBC, is narrow on its face and explicit in its limits. “I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model,” Lutnick wrote. The action allows the company to restore access to its Mythos 5 models to a list of about 100 “trusted partners,” including companies and federal agencies approved by the government, an administration official familiar with the move told Politico, in its write-up of the Friday letter partially reversing the export ban.

Anthropic welcomed the turnabout in a statement Friday night. “We received notice from the US government that Mythos 5, our strongest cybersecurity model, can be redeployed to a small group of cyber defenders and infrastructure providers,” the company said. “We are working to provision the approved set of providers and restore their access to Mythos 5 as quickly as possible. We are pleased to see this progress and continue to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again.”

The letter is silent on Fable 5. According to Politico, the recent letter does not mention any change to the government’s restrictions on use of the Fable 5 model, and conversations between Anthropic and the government are expected to continue into the weekend on Fable 5, a source familiar with the discussions told CNN. The Friday move is the opening of a partial door, not a full reopening.

I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model.

U.S. Commerce Secretary Howard Lutnick, in a Friday letter to Anthropic viewed by CNBC, which published the letter determining Mythos 5 safeguards are adequate.

Who Gets Access Back

The trusted-partner list is shorter than the model’s pre-ban reach, and tighter on purpose. About 100 organizations, a mix of federal agencies and private companies, will see Mythos 5 again, with the restoration aimed at defensive cyber purposes, according to people familiar with the matter. The narrower scope mirrors the audience Mythos 5 already had before the ban: a curated subset of defenders inside Anthropic’s Project Glasswing program. Anthropic says it is provisioning the approved set as quickly as possible.

The pre-ban Glasswing cohort already included some of the most security-sensitive names in US industry. NBC News, in its coverage of the partial restoration letter and Fable 5 silence, named the kinds of organizations that had Mythos 5 access before the export-control directive:

• Cisco, the infrastructure provider long involved in US network defense.
• JPMorgan Chase, the bank that had been running the model on internal security work.
• A broader Glasswing partner cohort of cyber defenders and critical-software maintainers vetted by Anthropic.

Anthropic has not published the post-ban roster, and the company declined to share which of its partners will get access first. The 100-name ceiling signals a tighter perimeter than the pre-ban Glasswing audience, not a return to it.

The Two Models Split Apart by Design

Fable 5 and Mythos 5 ship from the same underlying model, but they ship to different audiences with different guardrails. Both launched on June 9, 2026, Anthropic confirmed in its own announcement, with Mythos 5 framed as a stronger version of the same base for vetted defenders.

Fable 5 is wrapped in safety classifiers that watch the user’s prompt and route sensitive queries to a less-capable fallback, Claude Opus 4.8. Anthropic’s own announcement says those classifiers trigger, on average, in less than 5% of sessions, and cover cybersecurity, biology, chemistry, and model-distillation queries. “We’ve launched the model with safeguards that mean queries on some topics will instead receive a response from our next-most-capable model, Claude Opus 4.8,” Anthropic wrote in its launch announcement for both models.

Mythos 5 lifts those safeguards in some areas, which is what made it useful for cyber-defense research and what made it a national-security risk if it leaked. The same underlying intelligence gets exposed to defenders and, in the government’s view, to anyone who could extract it from the model’s behavior. That asymmetry is the entire reason a public-release ban applied to both models at once.

How the Two-Week Standoff Began

The standoff traces back to a single phone call on Friday, June 12, 2026. Commerce Secretary Howard Lutnick invoked export-control authorities under the Export Control Reform Act and ordered Anthropic to suspend all access by any foreign national to both Fable 5 and Mythos 5, inside or outside the United States.

The trigger was a workaround Amazon’s AI researchers had found on Fable 5. “Administration officials say the concerns were realized last week when Amazon AI experts were able to orchestrate a ‘jailbreak’ on Fable 5 after it was released on June 9,” Fox Business reported. Amazon CEO Andy Jassy reportedly alerted the White House directly. The administration’s reading was that Fable 5’s guardrails could be stripped away by anyone willing to invest the effort, which would expose Mythos-class cyber capability to the open internet.

Anthropic disputed the framing in a June 12 statement, saying the underlying vulnerabilities the workaround found were minor and previously known, and that other publicly available models could find them too. The company’s bigger objection was to the precedent. “If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers,” Anthropic wrote. With foreign-national employees inside the company and at partner organizations, the export-control regime forced Anthropic to disable both models for every customer worldwide, including US-based ones.

Anthropic dispatched a team of its top scientists and engineers to Washington to work with counterparts in the Commerce Department and the Office of the National Cyber Director. Two weeks of meetings followed. The Friday letter is the first concrete output of that engagement.

OpenAI’s GPT-5.6 Launched Under the Same Curbs

Hours before the Lutnick letter landed, OpenAI launched GPT-5.6 Sol, Terra, and Luna, three new models in its flagship family. At the administration’s request, OpenAI is releasing the models in phases, initially to a small group of trusted partners, with broader availability planned for the coming weeks.

OpenAI CEO Sam Altman called the staggered debut “bad news” in a post on X, since OpenAI had planned a wider, open-access launch. The company previewed the models’ capabilities and shared its trusted-partners list with the government ahead of Friday’s release.

The framing signals that the ad-hoc vetting process that just reopened Mythos 5 is becoming a routine step before any frontier model rollout in Washington. OpenAI, in a blog post Friday, said it would work with the Trump administration to develop a more robust framework to vet and clear future models for public release, while also flagging the cost of the current arrangement.

We don’t believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.

OpenAI, in its GPT-5.6 blog post, Friday.

The pattern is consistent across the two launches. OpenAI previewed capabilities, shared the trusted-partners list, and agreed to a phased release. Anthropic got the same playbook applied retroactively to a model that was already out. Both companies are now inside a vetting framework with no published rulebook, and both have reason to keep it from hardening into a license regime. For more on the OpenAI side, the company’s earlier deal with the White House on a smaller partner group is laid out in the 20-partner release OpenAI agreed to with the White House.

The Stakes for the Wider Frontier Rollout

The Lutnick letter settled one piece of a fight that is still open on several fronts. A group of nearly 80 cybersecurity executives and experts sent an open letter to Lutnick and National Cyber Director Sean Cairncross asking the White House to lift the restrictions and “commit to an open, scientific and transparent process of handling AI risk assessments in the future,” as the IAPP reported in its write-up of the global reaction to the Anthropic export directive. The letter warned that pulling the best capabilities away from defenders, while adversaries advance rapidly, is dangerous for US cyber posture.

Anthropic is also fighting the administration on a separate legal track. Earlier this year, the Department of Defense labeled Anthropic a supply chain risk to national security after disagreements about military uses for Anthropic’s models, a designation historically reserved for foreign adversaries. Anthropic sued the Trump administration to reverse the blacklisting, and litigation is still ongoing. The Friday letter does not touch that case, but it lands on top of a company that already had one foot in court with Washington.

The Friday decision also sits inside a federal review architecture that is still under construction. At the beginning of June, President Trump signed an executive order directing the federal government to shore up key cyber defenses and establish a mechanism for testing the most advanced AI models for safety issues. That mechanism remains in development. Until it ships, each new frontier model lands in the same ad-hoc space the Lutnick letter just negotiated. The Monday meeting between Anthropic and administration officials in Washington, which ended without a deal, is laid out in the read-out of Anthropic’s Washington meeting that kept Fable 5 blocked.

Fable 5 Remains the Open Question

Fable 5 is the model Anthropic needs back, and the one the Lutnick letter leaves offline. The Friday letter does not mention any change to restrictions on the public version of the model, and a source close to the company told NBC that Anthropic will continue to hold discussions with the government over the weekend as it seeks to restore Fable 5 to general use.

Commerce Department spokesperson Benno Kass framed the partial restoration as a first step. “In just two weeks, we have worked diligently to ensure America remains the global leader in AI while safeguarding our security,” Kass said. The unfinished business is Fable 5’s return to the paying customers and developers who got three days of access before the directive landed. Anthropic’s broader case for keeping the model available is documented in the company’s letter to senators on the Alibaba Claude distillation, a separate fight that ran alongside the export-control dispute. Fable 5’s restoration depends on whether Anthropic can convince the Commerce Department that the Fable 5 guardrails, not just Mythos 5’s, hold up under the same standard the administration just applied.

Frequently Asked Questions

What are Anthropic’s Mythos 5 and Fable 5?

Mythos 5 and Fable 5 are two products built on the same underlying Anthropic model, both launched June 9, 2026. Mythos 5 was released to a vetted group of cyber defenders and infrastructure providers through Project Glasswing. Fable 5 went to the general public but is wrapped in safety classifiers that route cyber, biology, chemistry, and model-distillation queries to Claude Opus 4.8.

Why did the US government block the models?

On June 12, 2026, the Commerce Department issued an export-control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, inside or outside the United States, citing national security concerns. The trigger was a workaround Amazon’s AI researchers reported on Fable 5 that the administration believed could strip away its guardrails.

What changed in Friday’s Lutnick letter?

Commerce Secretary Howard Lutnick determined that appropriate safeguards are in place for certain trusted partners to access Mythos 5. Anthropic can restore Mythos 5 access to about 100 federal agencies and private companies, focused on defensive cyber purposes. The letter does not address Fable 5.

What about OpenAI’s GPT-5.6?

OpenAI launched GPT-5.6 Sol, Terra, and Luna the same day, also limited at the administration’s request to a small group of trusted partners. OpenAI said it plans to make the models generally available in the coming weeks and is working with the Trump administration on a framework for vetting future frontier models.

Will Fable 5 come back?

Anthropic says weekend talks with the government continue toward restoring Fable 5 to general use. The Lutnick letter does not change Fable 5’s status, and discussions are expected to continue into the weekend. Fable 5’s restoration depends on whether the Commerce Department accepts that Fable 5’s guardrails hold up under the same standard the administration just applied to Mythos 5.